Association of Prenatal Exposure to Organophosphate, Pyrethroid, and Neonicotinoid Insecticides with Child Neurodevelopment at 2 Years of Age: A Prospective Cohort Study

Abstract

Background:

Widespread insecticide exposure might be a risk factor for neurodevelopment of our children, but few studies examined the mixture effect of maternal coexposure to organophosphate insecticides (OPPs), pyrethroids (PYRs), and neonicotinoid insecticides (NNIs) during pregnancy on child neurodevelopment, and critical windows of exposure are unknown.

Objectives:

We aimed to evaluate the association of prenatal exposure to multiple insecticides with children’s neurodevelopment and to identify critical windows of the exposure.

Methods:

Pregnant women were recruited into a prospective birth cohort study in Wuhan, China, from 2014–2017. Eight metabolites of OPPs (mOPPs), three metabolites of PYRs (mPYRs), and nine metabolites of NNIs (mNNIs) were measured in 3,123 urine samples collected at their first, second, and third trimesters. Children’s neurodevelopment [mental development index (MDI) and psychomotor development index (PDI)] was assessed using the Bayley Scales of Infant Development at 2 years of age ($N=\mathrm{1,041}$). Multivariate linear regression models, generalized estimating equation models, and weighted quantile sum (WQS) regression were used to estimate the association between the insecticide metabolites and Bayley scores. Potential sex-specific associations were also examined.

Results:

Single chemical analysis suggested higher urinary concentrations of some insecticide metabolites at the first trimester were significantly associated with lower MDI and PDI scores, and the associations were more prominent among boys. Each 1-unit increase in ln-transformed urinary concentrations of two mOPPs, 3,5,6-trichloro-2-pyridinol and 4-nitrophenol, was associated with a decrease of 3.16 points [95% confidence interval (CI): $-5.59$, $-0.74$] and 3.06 points (95% CI: $-5.45$, $-0.68$) respectively in boys’ MDI scores. Each 1-unit increase in that of trans-3-(2,2-dichloroethenyl)-2,2-dimethylcyclopropanecarboxylic acid (trans-DCCA; an mPYR) was significantly associated with a decrease of 2.24 points (95% CI: $-3.89$, $-0.58$) in boys’ MDI scores and 1.90 points (95% CI: $-3.16$, $-0.64$) in boys’ PDI scores, respectively. Significantly positive associations of maternal urinary biomarker concentrations [e.g., dimethyl phosphate (a nonspecific mOPP) and desmethyl-clothianidin (a relatively specific mNNI)] with child neurodevelopment were also observed. Using repeated holdout validation, a 1-quartile increase in the WQS index of the insecticide mixture (in the negative direction) at the first trimester was significantly associated with a decrease of 3.02 points (95% CI: $-5.47$, $-0.57$) in MDI scores among the boys, and trans-DCCA contributed the most to the association (18%).

Conclusions:

Prenatal exposure to higher levels of certain insecticides and their mixture were associated with lower Bayley scores in children, particularly in boys. Early pregnancy may be a sensitive window for such an effect. Future studies are needed to confirm our findings. https://doi.org/10.1289/EHP12097

Introduction

Maternal exposure to environmental chemicals during pregnancy might impair fetal development and increase the future health risks of the children.^1–4 Organophosphates (OPPs), pyrethroids (PYRs), and neonicotinoids (NNIs) are three major classes of insecticides widely used in the world.^5,6 Neurotoxicity of prenatal exposure to these insecticides has been documented in animal models.^7–10 Increasing evidence, mainly from populations with elevated exposure as a result of living near agricultural areas, indicated that prenatal exposure to OPPs,^11–14 PYRs,^15,16 or NNIs may impair child neurodevelopment,^17,18 causing adverse outcomes, such as decreased Bayley scores, or increased risk of autism spectrum disorder (ASD).

Previous studies have investigated neurocognitive development in children associated with gestational exposure to OPPs^19–23 or PYRs^24,25 or both of them²⁶ in general populations. The available results are inconsistent; some studies have observed harmful effects of OPPs^19,20,22 or PYRs,²⁴ including sex-specific effects,^21,23 whereas some others found no significant associations²⁵ or positive effects.²⁶ On the other hand, NNIs once accounted for $>25\%$ of the global insecticide market²⁷; however, epidemiological studies on the neurodevelopment effects of NNIs in the general population, particularly based on biomonitoring data, are still scarce.

The general population can be exposed to multiple insecticides simultaneously through the intake of contaminated food^28,29 and drinking water,³⁰ but few epidemiological studies have explored the mixture effect of maternal exposure to multiple classes of insecticides during pregnancy on child neurodevelopment. In addition, many of those measured exposure biomarkers in urine at only one time point.^20,23,26,31 However, for those insecticides that have relatively short half-lives in the human body,^32,33 repeated measurements of urinary biomarkers across pregnancy (the three trimesters) are strongly desirable to increase the reliability of exposure assessment and to examine potential sensitive windows for adverse effects of prenatal exposure in relation to neurodevelopment.

In this study, exposure biomarker concentrations of OPPs, PYRs, and NNIs were measured in urine samples repeatedly collected from 1,041 pregnant Chinese women during the first, second, and third trimesters from Wuhan, China, between 2014 and 2017. We explored whether there was a potential sensitive window of prenatal insecticide exposure in association with child neurodevelopment. Because previous studies suggested that prenatal insecticide exposure has stronger adverse effects on neurodevelopment in male offspring,^34,35 sex-stratified analyses were performed in this study, and the interaction term between insecticide exposure and child sex was then used to examine the effect modification. Mixture effect of exposure to the selected insecticides on child neurodevelopment was explored and the primary contributors were identified.

Methods

Study Population

Participants in this study were part of a prospective birth cohort that aimed to investigate the association of environmental exposures with the health of pregnant women and their children who were recruited at the first antenatal examination ($<16\mathrm{wk}$ of gestation) in the Wuhan Women’s and Children’s Health Care Center, Wuhan, central China. Participants were considered eligible if they were $>18$ years of age, were Wuhan residents (lived in Wuhan for $\ge 1\mathrm{y}$) who could understand Chinese well and complete the questionnaire independently without communication problems (i.e., no mental illness), had a singleton pregnancy, planned to deliver at the study hospital, and were willing to provide biospecimens at three trimesters during pregnancy. All participants in this study were urban residents (determined by their permanent address) with no occupational exposure to pesticides. The research protocol was authorized by the ethics committees of Tongji Medical College and the participating health care center, and all the participants provided informed consent.

From January 2014 to June 2017, a total of 5,112 pregnant women were recruited into the cohort and donated at least one urine sample, of which 2,782 mothers had their children complete the neurodevelopment assessment at 2 years of age through May 2020. Of the 2,782 mother–child pairs, 1,041 mothers who donated one spot urine sample in each of the three trimesters during gestation (i.e., who had all three urine samples) were included in this study. The rest, who provided fewer than three urine samples, were not included in this study. No statistically significant differences for the demographic characteristics were observed among the recruited population ($N=\mathrm{5,112}$), the women whose children completing the evaluation of Bayley scales at 2 years of age ($N=\mathrm{2,782}$), or the study population ($N=\mathrm{1,041}$) (Table S1).

Data Collection

The data of demographic and socioeconomic characteristics (maternal age, prepregnancy weight, annual household income, and both maternal and paternal education) and lifestyle factors (smoking, passive smoking, alcohol use, and folic acid supplementation during pregnancy) were obtained in face-to-face interviews at enrollment or during prenatal follow-up visits by trained nurses using standardized questionnaires. The data about pregnancy complications (maternal anemia, hypertensive disorders in pregnancy, and gestational diabetes mellitus) and the information of newborns (child sex, gestational age at birth, birth weight, and birth height) were retrieved from the medical records at birth. Breastfeeding status was obtained from questionnaires during the follow-up of children. In addition, self-reported prepregnancy weight and height were applied to calculate the prepregnancy body mass index (PBMI), which was categorized into three groups according to the Guideline of Chinese adults: underweight ($<18.5$), normal weight ($18.5–23.9$), and overweight ($\ge 24{\mathrm{kg}/\mathrm{m}}^2$). Gestational weight gain (GWG; in kilograms) was divided into three groups: inadequate, recommended, and excessive weight gain during pregnancy (Table S2) according to the standard of recommendation of the Chinese National Health Commission (NHC).³⁶ Sampling seasons were obtained according to the sampling date with May–October and November–April, respectively representing the warm and cold seasons in Wuhan.

Assessment of Child Neurocognitive Development

Neurocognitive development for each child was assessed at 2 years of age (range: 23–26 months) using the locally standardized Chinese revision of the Bayley Scales of Infant Development (BSID-CR).^37,38 Cognition, language, and social development was assessed using the mental development index (MDI); fine and gross motor development was assessed using the psychomotor development index (PDI).³⁸

Two well-trained certified psychologists administered the BSID-CR in a quiet room of the study health care center. Videotaped evaluations were used for quality control. Cronbach’s alpha intraclass correlation coefficient³⁹ for the 2-year-old BSID-CR assessments between two psychologists was 0.99 ($p<0.001$) on the basis of double-scoring among 5% of the randomly selected participants.

Urinary Biomarker Measurements and Quality Control

For each pregnant woman, one spot urine sample was collected at the first ($13.0\pm 1.0\mathrm{wk}$), second ($24.5\pm 3.6\mathrm{wk}$), and third trimesters ($33.9\pm 2.9\mathrm{wk}$) during pregnancy, respectively. The samples were stored in polypropylene tubes at $-20°\mathrm{C}$ for further analysis.

In brief, based on previous studies, we measured six nonspecific metabolites of OPPs [i.e., dialkyl phosphates (DAPs), including dimethyl phosphate (DMP), dimethyl thiophosphate (DMTP), dimethyl dithiophosphate (DMDTP), diethyl phosphate (DEP), diethyl thiophosphate (DETP), diethyl dithiophosphate (DEDTP)], two specific metabolites of OPPs [3,5,6-trichloro-2-pyridinol (TCPy) and 4-nitrophenol (PNP)],^40,41 three typical metabolites of PYRs [trans-3-(2,2-dichlorovinyl)-2,2-dimethylcyclopropanecarboxylic acid (trans-DCCA), 3-phenoxybenzoic acid (3-PBA), and 4-fluoro-3-phenoxybenzoic acid (4F-3-PBA)],⁴⁰ and nine typical NNIs/their metabolites [imidacloprid (IMI), acetamiprid (ACE), thiamethoxam (THM), clothianidin (CLO), 5-hydroxy-imidacloprid (5-hydroxy-IMI), imidacloprid-olefin (IMI-olefin), desnitro-imidacloprid (DN-IMI), desmethyl-acetamiprid (DM-ACE), and desmethyl-clothianidin (DM-CLO)].⁴² The eight metabolites of OPPs (defined as mOPPs), three metabolites of PYRs (defined as mPYRs), and nine NNIs/their metabolites (defined as mNNIs) were measured using isotope-dilution mass spectrometry (MS) methods (Table S3).

Details on specific sample pretreatment steps are as follows. After thawing, $1\mathrm{mL}$ of each urine sample was transferred to a $15\text{-mL}$ centrifuge tube (Corning Inc.), spiked with 850 U of $\mathrm{\beta }\text{-glucuronidase}$ ($100\mathrm{\mu }\mathrm{L}$ of $\mathrm{8,500}\mathrm{U}/\mathrm{mL}$ in $1\mathrm{M}$ ammonium acetate with $\text{pH}=5.0$) and internal standard mixture ($10\mathrm{\mu }\mathrm{L}$ of $100\mathrm{\mu }\mathrm{g}/\mathrm{L}$ in acetonitrile: $\text{water}=1$:1), and then incubated (shaken gently) at 37°C overnight. Then, $3\mathrm{mL}$ of ethyl acetate was added to each sample for extraction, vortexed in an automatic vortex shaker for 15 min, and centrifuged at $\mathrm{2,300}\times g$ for 10 min. The extraction step was repeated. The supernatants were combined and transferred into a glass tube, and then dried under a gentle nitrogen stream at 35°C. Finally, the sample was reconstituted in $0.5\mathrm{mL}$ of acetonitrile/water (3:7) and transferred into a $2\text{-mL}$ amber liquid chromatography (LC) vial, and stored at $-20°\mathrm{C}$ for further instrumental analysis.

Because DAPs and DN-IMI cannot be well extracted by the above liquid–liquid extraction method and because PNP had a relatively high background contamination during liquid–liquid extraction, we used the following sample preparation method for those analytes. In brief, $0.1\mathrm{mL}$ of each urine sample was spiked with 85 U of $\mathrm{\beta }\text{-glucuronidase}$ (dissolved in $10\mathrm{\mu }\mathrm{L}$ of $1\mathrm{M}$ ammonium acetate buffer, $\text{pH}=5.0$) and $1\text{\hspace{0.17em}ng}$ (for ${\text{DMP-}}^{13}{\mathrm{C}}_2$, DMTP-d6, ${\text{DMDTP-}}^{13}{\mathrm{C}}_2$, ${\text{DEP-}}^{13}{\mathrm{C}}_4$, ${\text{DETP}\text{-d}}_{10}$, ${\text{DEDTP-}}^{13}{\mathrm{C}}_4$, ${\text{PNP-}}^{13}{\mathrm{C}}_6$, and $\text{DN-IMI-}{\text{d}}_4$) of each isotope-labeled internal standard, and incubated at 37°C overnight. Then, the sample was diluted five times with $400\mathrm{\mu }\mathrm{L}$ of 0.05% formic acid in water, vortexed, transferred into an Amicon Ultra-0.5 Centrifugal Filter Unit,⁴³ and centrifuged at $\mathrm{12,000}\times g$ for 30 min. The filtered sample was transferred into an amber LC vial for determination of the target analytes.

Urinary target analyte concentrations were measured by an ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS; ExionLC-QTRAP $6500+$, AB SCIEX) with an ACQUITY UPLC HSS T3 column ($1.8\mathrm{\mu }\mathrm{m}$, $2.1\mathrm{mm}\times 150\mathrm{mm}$; Waters Corporation) maintained at 40°C. Details about the multiple reaction monitoring (MRM) transitions of the target analytes are presented in Table S3. The details on the gradient elution program, mobile phases used, and flow rate are shown in Tables S4 and S5. The separation of the target analytes except for DAPs, PNP, and DN-IMI was achieved with a gradient elution program (Table S4) at a flow rate of $0.3\mathrm{mL}/\text{min}$. For DAPs, PNP, and DN-IMI, the target analytes were determined separately using another gradient elution program (Table S5) at a flow rate of $0.25\mathrm{mL}/\text{min}$. The injection volume was $10\mathrm{\mu }\mathrm{L}$. The detection was conducted in MRM mode. mOPPs, IMI-olefin, and DM-CLO were monitored in negative electrospray ionization (${\text{ESI}}^{-}$) mode with the ionization voltage of $-\mathrm{4,500}\mathrm{V}$. The rest target analytes were monitored in ${\text{ESI}}^{+}$ with an ionization voltage of $\mathrm{5,500}\mathrm{V}$. The ion source temperature was 650°C.

Details on the quality assurance/control are as follows: the calibration curves of the analytes were obtained based on 12 levels (0.002, 0.005, 0.01, 0.02, 0.05, 0.10, 0.20, 0.50, 1.00, 2.00, 5.00, and $10.0\text{\hspace{0.17em}ng}/\mathrm{mL}$, with $2\text{\hspace{0.17em}ng}/\mathrm{mL}$ of each internal standard), and the regression coefficient ($R^2$) was $\ge 0.999$ for each target analyte. Procedural blanks, travel blanks, duplicates (two pooled samples as quality control materials), matrix spikes, and checks of carry-over and instrumental sensitivity drift were included in each batch. The average recoveries of matrix spiked native standards ($1\text{\hspace{0.17em}ng}$ of each native standard in the urine sample) ranged from 86.0% to 115%. The relative standard deviations (SDs) of duplicates were $<10.0\%$. Instrumental limit of quantifications (LOQs) and method detection limits (MDLs) (shown in Table S3) have been defined elsewhere.⁴⁴ The LOQs of the analytes ranged from 0.005 to $0.20\text{\hspace{0.17em}ng}/\mathrm{mL}$ (Table S3). The target analytes were not found in any of the blanks or found to be less than the LOQs.

Creatinine adjustment may not be appropriate for metabolite levels in populations undergoing rapid physiologic changes, such as pregnant women, owing to high intra-individual variability in creatinine excretion.⁴⁵ It has been demonstrated that creatinine can be significantly influenced by age, BMI, and fat-free mass,⁴⁶ whereas, specific gravity (SG) could be more reliable compared with creatinine.⁴⁷ Thus, in this study, SG of urine was used to correct urine dilution.^44,48 Specifically, the urinary SG was measured by a digital handheld refractometer (PAL-10S, Atago). Concentrations of target pesticides were standardized by urinary SG using the following equation: $C_{SG}=C\times \left[\left(S{G}_m-1\right)/\left(SG-1\right)\right]$, where $C_{SG}$ is the SG-adjusted concentration (in nanograms per milliliter), C is the observed concentration (in nanograms per milliliter), and $S{G}_m$ is the median SG at each trimester.

Statistical Analysis

Descriptive statistics on characteristics of the mothers and children were performed, and the chi-square test was conducted to examine the differences in basic characteristics of mother–child pairs among the recruitment population ($N=\mathrm{5,112}$), the population of their children completing the evaluation of Bayley scales at 2 years of age ($N=\mathrm{2,782}$), and the study population ($N=\mathrm{1,041}$). The analysis of variance was used to examine whether child MDI and PDI scores at 2 years of age differed among the participants with different characteristics. Descriptive statistics were also used to characterize the distribution of child Bayley scores at 2 years of age based on the quartiles of average urinary concentrations of insecticide biomarkers throughout pregnancy for the participants.

DMDTP, DEDTP, 4F-3-PBA, and ACE were excluded in the following statistical analysis owing to low detection frequencies (DFs, $<30\%$). The other mOPPs, mPYRs, and mNNIs had DFs of $>70\%$, and values below the MDLs were imputed as the MDL divided by the square root of 2. To approximate normal distributions, all concentrations were natural logarithm (ln)-transformed for further statistical analysis. We examined Spearman’s rank correlations between target analytes at individual time points and on the basis of the averaged prenatal exposure (the arithmetic mean concentrations of mOPPs, mPYRs, and mNNIs at the three trimesters). Different from many previous studies, we did not calculate the summed concentrations of six nonspecific metabolites of OPPs because DMP and DEP can be originated from not only OPPs but also the organophosphate esters trimethyl phosphate (TMP)⁴⁹ and triethyl phosphate (TEP),⁵⁰ which might cause overestimation of the total exposure to OPPs. TMP and TEP can be metabolized into DMP and DEP and excreted through the urine,⁵¹ and both TMP and TEP are ubiquitous in the environment.⁵² In addition, we also compared the concentrations of mOPPs and mPYRs observed in the present study with those in adult females ($>19\text{\hspace{0.17em}years}$ of age) from the U.S. National Health and Nutrition Examination Survey (NHANES) in 2013–2014 ($N=790$)⁵³ and in 2017–2018 ($N=864$).⁵⁴

To investigate the effects of insecticide exposure during pregnancy on child neurodevelopment, multivariate linear regression (MLR) models were performed to evaluate the associations between the averaged concentrations of individual analytes across three trimesters and children’s MDI/PDI scores at 2 years of age. Moreover, to uncover the critical windows of exposure susceptibility during pregnancy, generalized estimating equation (GEE) models were used to assess trimester-specific associations of target analyte concentrations with children’s MDI/PDI scores.⁵⁵ Considering that pesticides may have sex-specific effects, we performed sex-stratified analyses to obtain the effect estimates ($\mathrm{\beta }$) and 95% confidence intervals (95% CI), and the corresponding $p$-values. In addition, to examine the effect modification by child sex, we also introduced the interaction term ($\text{child\hspace{0.17em}sex}\times \text{urinary\hspace{0.17em}level\hspace{0.17em}of\hspace{0.17em}insecticide\hspace{0.17em}metabolite}$) into to the models (whole population) to derive interaction $p$-values (defined as $p_{\text{sex}\text{-int}}$). The false discovery rate (FDR) corrections were performed on $p$-values ($p_{FDR}$) to account for multiple comparisons.⁵⁶

Weighted quantile sum (WQS) regression (WQSR) was conducted to identify the major contributors to the association.⁵⁷ WQSR analysis is an approach that can be used in environmental epidemiology to evaluate associations between potentially highly correlated coexposures and a health outcome, thereby reducing dimensionality and avoiding multicollinearity.⁵⁸ The WQSR coefficient can be interpreted as the mixture effect of the insecticide mixture exposure on the outcome. The average weight for each biomarker, expressed as the percentage of that sum to one, was estimated from 1,000 bootstraps. Based on the results of MLR and GEE models that suggested the susceptible population and windows of insecticide exposure for adverse effects on neurodevelopment, WQSR analysis was further conducted in the negative direction of the outcomes to identify which biomarker contributed more to the adverse mixture effect. To evaluate the robustness of our estimates, we applied repeated holdout validation in WQSR models.⁵⁹ Specifically, the full data set was randomly split into 40%/60% training–testing sets and the WQSR analysis was repeated 100 times to examine the distribution of effect estimates and associated weights.⁵⁹

Covariates were considered as potential confounders if they altered the estimate of the association by 10% or more by a step-wise backward elimination approach,⁶⁰ including maternal age, PBMI, maternal education level, paternal education level, parity, delivery mode, passive smoking during pregnancy, child sex, and breastfeeding duration (Excel Table S1). In addition, folic acid supplementation during pregnancy⁶¹ and season of sample collection⁴² were also included, considering that these variables have been reported to be significantly associated with child neurodevelopment and insecticide exposure, respectively. Among those covariates above, a multicollinearity between maternal education level and paternal education level was observed in the regression model [$\text{variance\hspace{0.17em}inflation\hspace{0.17em}factor\hspace{0.17em}}\left(\text{VIF}\right)=13.0$], given that VIF values $>10$ indicate serious multicollinearity. Thus, only maternal education level was included as a covariate in the final models because its adjustment showed a larger impact on the estimates for the associations between some insecticide biomarkers (i.e., DMTP, TCPy, PNP, 3-PBA, trans-DCCA, IMI, IMI-olefin, THM, and DM-CLO) and Bayley scores than adjustment for paternal education level (Excel Table S1). The final models were adjusted for maternal education (categorical: $\le \text{high\hspace{0.17em}school}$, bachelor’s degree, or $\ge \text{master}’\mathrm{s}\text{\hspace{0.17em}degree}$), maternal age (categorical: $<25$, 25–29, 30–34, or $\ge 35\mathrm{y}$), PBMI (categorical: $<18.5$, 18.5–23.9, or $\ge 24{\mathrm{kg}/\mathrm{m}}^2$), folic acid supplementation during pregnancy (categorical: yes or no), passive smoking during pregnancy (categorical: yes or no), delivery mode (categorical: vaginal delivery or cesarean delivery), child sex (categorical: boys or girls), parity (categorical: nulliparous or multiparous), breastfeeding duration (categorical: $<6$ or $\ge 6\text{\hspace{0.17em}months}$), and season of sample collection [categorical: warm season or cold season (except for the average across all trimesters)]. In the present analyses, only breastfeeding duration had missing data (2.9% of values), which were imputed by randomly selecting a value from participants with nonmissing values¹¹ to preserve the size of the study population.

In addition, to investigate whether adverse perinatal factors affected the reliability of the results, we performed three sensitivity analyses based on MLR models to test the robustness of the associations with

1. Additional adjustment for more covariates—which were related but altered the estimate of the association by $<10\%$, including household income (categorical: $<\mathrm{50,000}$, 50,000–100,000, or $\ge \mathrm{100,000}\text{\hspace{0.17em}CNY}$), nutritional status during pregnancy [assessed by maternal anemia (categorical: yes or no)], pregnancy weight gain (categorical: inadequate total GWG, adequate total GWG, or excessive total GWG), birth weight (categorical: $<\mathrm{2,500}$, 2,500–4,000, or $>\mathrm{4,000}\mathrm{g}$), and paternal education level (categorical: $\le \text{high\hspace{0.17em}school}$, bachelor’s degree, or $\ge \text{master}’\mathrm{s}\text{\hspace{0.17em}degree}$)—to examine the influence of these factors potentially on the associations

2. Excluding women with gestational diabetes or gestational hypertension

3. Excluding children with preterm birth (gestation age $<37\mathrm{wk}$) or low birth weight ($<\mathrm{2,500}\mathrm{g}$).

All statistical analyses were conducted using SAS (version 9.4; SAS Institute, Inc.) or the gWQS package of R (version 4.0.4; R Development Core Team). The threshold for statistical significance was set at $p<0.05$ (two-tailed) or $p_{FDR}<0.05$.

Results

Participant Characteristics

As shown in Table 1, the average maternal age at delivery was $29.4\pm 3.7\mathrm{y}$. Of the 1,041 mothers, the majority were nulliparous (77.4%, $n=806$), well-educated (79.4% had a bachelor’s degree or higher, $n=827$), had a normal PBMI ($18.5–23.9{\mathrm{kg}/\mathrm{m}}^2$; 66.7%, $n=694$), had no anemia (96.0%, $n=999$), and reported folic acid supplementation (87.6%, $n=912$). Approximately 51.3% ($n=534$) of the mothers delivered by cesarean delivery, 42.9% ($n=447$) had excessive total GWG, and 25.1% ($n=261$) reported passive smoking during pregnancy. Among their children, 53.1% ($n=553$) were boys, 56.7% ($n=574$) were breastfed for $>6\text{\hspace{0.17em}months}$, and 91.8% ($n=956$) had a normal birth weight ($\mathrm{2,500}–\mathrm{4,000}\mathrm{g}$). The average ($\pm \text{SD}$) gestational age at delivery was $39.4\pm 1.1\mathrm{wk}$. Of the 3,123 urine samples, 49.7% ($n=\mathrm{1,552}$) were collected in the warm season (from May to October).

Table 1.

Descriptive characteristics of study participants and child neurodevelopment measures (Bayley scores at 2 years of age) from a birth cohort study in Wuhan, China, 2014–2017 ($N=\mathrm{1,041}$ mother–child pairs).

Characteristic	$N$ (%) or $\text{mean}\pm \text{SD}$	MDI scores		PDI scores
		$\text{Mean}\pm \text{SD}$	$p$-Value	$\text{Mean}\pm \text{SD}$	$p$-Value
Total	1,041	$110\pm 21.8$	—	$111\pm 17.1$	—
Maternal characteristics
Age at delivery (y)	$29.4\pm 3.7$	—	0.24	—	0.25
$<25$	60 (5.80)	$106\pm 19.5$	—	$114\pm 15.1$	—
25–29	548 (52.6)	$110\pm 22.2$	—	$110\pm 17.0$	—
30–34	330 (31.7)	$111\pm 21.3$	—	$112\pm 17.7$	—
$\ge 35$	103 (9.90)	$107\pm 21.8$	—	$109\pm 16.1$	—
PBMI (${\mathrm{kg}/\mathrm{m}}^2$)	$21.1\pm 2.91$	—	0.14	—	0.90
$<18.5$	195 (18.7)	$111\pm 20.4$	—	$111\pm 16.6$	—
18.5–23.9	694 (66.7)	$110\pm 22.0$	—	$111\pm 17.3$	—
$\ge 24$	152 (14.6)	$107\pm 22.1$	—	$111\pm 16.8$	—
Maternal education level	—	—	$<0.0001$	—	0.36
$\le \text{High-school degree}$	214 (20.6)	$104\pm 21.7$	—	$110\pm 17.6$	—
Bachelor’s degree	763 (73.3)	$111\pm 22.0$	—	$111\pm 16.9$	—
$\ge \text{Master}’\mathrm{s}\text{\hspace{0.17em}\hspace{0.17em}degree}$	64 (6.1)	$118\pm 14.6$	—	$114\pm 17.4$	—
Household income (CNY)	—	—	0.07	—	0.20
$<\mathrm{50,000}$	120 (11.5)	$109\pm 23.4$	—	$110\pm 15.7$	—
50,000–100,000	384 (36.9)	$108\pm 21.7$	—	$110\pm 18.0$	—
$\ge \mathrm{100,000}$	537 (51.6)	$111\pm 21.2$	—	$112\pm 16.7$	—
Parity	—	—	0.007	—	0.05
Nulliparous	806 (77.4)	$111\pm 21.8$	—	$110\pm 16.9$	—
Multiparous	235 (22.6)	$106\pm 21.4$	—	$113\pm 17.6$	—
Delivery mode	—	—	$<0.0001$	—	0.007
Vaginal delivery	507 (48.7)	$113\pm 22.0$	—	$112\pm 17.0$	—
Cesarean delivery	534 (51.3)	$107\pm 22.7$	—	$110\pm 17.0$	—
GWG categories according to NHC	—	—	0.88	—	0.77
Inadequate total GWG	165 (15.9)	$111\pm 22.0$	—	$110\pm 17.7$	—
Adequate total GWG	429 (41.2)	$110\pm 22.0$	—	$111\pm 17.7$	—
Excessive total GWG	447 (42.9)	$110\pm 21.5$	—	$111\pm 16.2$	—
Passive smoking during pregnancy	—	—	0.37	—	0.88
Yes	261 (25.1)	$109\pm 22.2$	—	$111\pm 18.6$	—
No	780 (74.9)	$110\pm 21.6$	—	$111\pm 16.4$	—
Folic acid supplementation during pregnancy	—	—	0.51	—	0.97
Yes	912 (87.6)	$110\pm 21.9$	—	$111\pm 17.1$	—
No	129 (12.4)	$109\pm 21.2$	—	$111\pm 16.5$	—
Maternal anemia	—	—	0.71	—	0.65
Yes	42 (4.0)	$109\pm 21.4$	—	$110\pm 15.8$	—
No	999 (96.0)	$110\pm 21.8$	—	$111\pm 17.1$	—
Hypertension in pregnancy	—	—	0.83	—	0.28
Yes	24 (2.30)	$109\pm 23.4$	—	$107\pm 15.5$	—
No	1,017 (97.7)	$110\pm 21.7$	—	$111\pm 17.1$	—
Gestational diabetes	—	—	0.35	—	0.70
Yes	98 (9.40)	$108\pm 23.1$	—	$112\pm 18.1$	—
No	943 (90.6)	$110\pm 21.6$	—	$111\pm 17.0$	—
Sampling season
First trimester	—	—	0.32	—	0.003
Warm season	523 (50.2)	$111\pm 21.8$	—	$109\pm 17.5$	—
Cold season	518 (49.8)	$109\pm 21.7$	—	$113\pm 16.4$	—
Second trimester	—	—	0.41	—	0.69
Warm season	499 (47.9)	$110\pm 22.2$	—	$111\pm 17.6$	—
Cold season	542 (52.1)	$109\pm 21.4$	—	$111\pm 16.6$	—
Third trimester	—	—	0.50	—	0.001
Warm season	530 (50.9)	$110\pm 21.7$	—	$113\pm 17.6$	—
Cold season	511 (49.1)	$109\pm 21.9$	—	$109\pm 16.3$	—
Paternal education level	—	—	0.004	—	0.49
$\le \text{High-school degree}$	209 (20.1)	$106\pm 22.5$	—	$110\pm 18.1$	—
Bachelor’s degree	783 (75.2)	$111\pm 21.6$	—	$111\pm 16.8$	—
$\ge \text{Master}’\mathrm{s}\text{\hspace{0.17em}\hspace{0.17em}degree}$	49 (4.70)	$114\pm 19.5$	—	$113\pm 16.2$	—
Child characteristics
Child’s sex	—	—	$<0.0001$	—	0.11
Boy	553 (53.1)	$107\pm 23.3$	—	$110\pm 17.2$	—
Girl	488 (46.9)	$113\pm 19.4$	—	$112\pm 16.9$	—
Birth weight (g)	$\mathrm{3,355}\pm 424$	—	0.23	—	0.38
$<\mathrm{2,500}$ (low birth weight)	21 (2.02)	$107\pm 26.0$	—	$106\pm 15.1$	—
2,500–4,000	956 (91.8)	$110\pm 21.5$	—	$111\pm 17.2$	—
$>\mathrm{4,000}$	64 (6.15)	$106\pm 24.0$	—	$111\pm 16.2$	—
Gestational age (wk)	$39.4\pm 1.1$	—	0.62	—	0.39
$<37$ (preterm birth)	28 (2.70)	$112\pm 19.1$	—	$109\pm 11.6$	—
$\ge 37$	1,013 (97.3)	$110\pm 21.8$	—	$111\pm 17.2$	—
Breastfeeding duration (months)	—	—	0.04	—	0.41
$<6$	437 (43.2)	$108\pm 22.1$	—	$112\pm 17.3$	—
$\ge 6$	574 (56.7)	$111\pm 21.5$	—	$110\pm 16.8$	—
Missing	30	—	—	—	—

Note: Values are $\text{mean}\pm \text{SD}$ or numbers (percentage). —, Not applicable; %, percentage; cold season, November–April; GWG, gestational weight gain; MDI, mental development index; NHC, National Health Commission of the People’s Republic of China; PBMI, prepregnancy body mass index; PDI, psychomotor development index; SD, standard deviation; SD, standard deviation; warm season, May–October. *$p$-Values were derived from the analysis of variance performed to compare MDI and PDI among different characteristic categories. $p$-Values of $<0.05$ were considered statistically significant.

The mean BSID-CR MDI and PDI scores ($\text{mean}\pm \text{SD}$) were respectively $110\pm 21.8$ and $111\pm 17.1$ for the children (Table 1). Higher MDI scores were observed among children who were breastfed for $\ge 6$ months ($p=0.04$) or whose mothers were nulliparous ($p=0.007$), had a vaginal delivery ($p<0.001$), or had a master’s degree ($p<0.001$). Girls’ MDI scores were significantly higher than boys’ MDI scores ($\text{mean}=113$ vs. 107, $p<0.001$). Lower PDI scores were observed among children whose mothers’ urine samples of the 1st ($p=0.003$) collected in the warm season (vs. cold season) and those of the 3rd trimester ($p=0.001$) collected in the cold season (vs. warm season). Distribution of Bayley scores in this study is shown in Table 2, and the Bayley scores ($\text{mean}\pm \text{SD}$) of participating children according to the averaged urinary quartile concentrations of insecticide biomarkers throughout pregnancy are shown in Table S6.

Table 2.

Distribution of Bayley Scales for Infant Development–Chinese revision (BSID-CR) scores (points) among child participants at 2 years of age from a birth cohort study in Wuhan, China ($n=\mathrm{1,041}$ participants).

Development index	$\text{Mean}\pm \text{SD}$	25th	50th	75th	95th	Minimum–maximum
MDI	$110\pm 21.8$	98	114	126	138	50–150
PDI	$111\pm 17.1$	99	111	123	140	51–150

Note: 25th, 50th, 75th, 95th mean the 25th, 50th, 75th, and 95th percentile scores, respectively. MDI, mental development index; PDI, psychomotor development index; SD, standard deviation.

Levels of Maternal Urinary mOPPs, mPYRs, and mNNIs

As shown in Table 3, most insecticide metabolites were detected in $>90\%$ of the maternal urine samples, except for IMI (DFs: 72.1%–74.5%) and DN-IMI (DFs: 82.0%–84.2%). Most mOPPs (range of their median SG-adjusted concentrations: $0.42–4.91\text{\hspace{0.17em}ng}/\mathrm{mL}$) had higher concentrations than mPYRs ($0.14–0.23\text{\hspace{0.17em}ng}/\mathrm{mL}$) and mNNIs ($0.04–1.07\text{\hspace{0.17em}ng}/\mathrm{mL}$).

Table 3.

Distribution of maternal urinary concentrations (ng/mL) of mOPPs, mPYRs, and mNNIs in participants from a birth cohort study in Wuhan, China, 2014–2017 ($N=\mathrm{1,041}$ participants, $n=\mathrm{1,041}$ samples for each trimester).

Compounds ($\text{ng}/\mathrm{mL}$)	MDL ($\text{ng}/\mathrm{mL}$)	Average trimester （$n=\mathrm{1,041}$）		First trimester			Second trimester			Third trimester
		GM	Median (25th–75th)	DF [percentage $>\text{MDL}$ (%)]	GM	Median (25th–75th)	DF [percentage $>\text{MDL}$ (%)]	GM	Median (25th–75th)	DF [percentage $>\text{MDL}$ (%)]	GM	Median (25th–75th)
mOPPs
DEP
Unadjusted	0.25	4.07	3.98 (2.53–6.32)	100	3.80	3.75 (1.99–6.80)	100	3.13	3.23 (1.63–5.63)	100	2.85	2.70 (1.48–5.59)
SG-adjusted	—	4.16	4.01 (2.63–6.20)	—	4.08	3.88 (2.30–6.54)	—	3.57	3.47 (2.01–6.11)	—	3.48	3.27 (2.04–5.96)
DMP
Unadjusted	0.25	4.97	4.79 (2.97–7.78)	100	4.83	4.74 (2.45–9.55)	100	3.41	3.26 (1.72–6.68)	100	3.12	2.99 (1.57–5.84)
SG-adjusted	—	5.07	4.78 (3.24–7.61)	—	5.18	4.91 (2.87–8.86)	—	3.89	3.57 (2.10–6.93)	—	3.82	3.79 (2.16–6.83)
DETP
Unadjusted	0.05	1.90	1.88 (1.11–3.29)	99.9	1.79	1.72 (0.85–3.68)	99.9	1.27	1.23 (0.64–2.60)	99.8	1.20	1.18 (0.55–2.54)
SG-adjusted	—	1.98	1.90 (1.16–3.14)	—	2.08	2.03 (1.09–3.69)	—	1.33	1.28 (0.66–2.54)	—	1.23	1.21 (0.63–2.41)
DMTP
Unadjusted	0.25	0.63	0.57 (0.34–1.07)	95.6	0.53	0.46 (0.26–1.00)	93.9	0.41	0.38 (0.25–0.81)	94.7	0.42	0.38 (0.25–0.76)
SG-adjusted	—	0.66	0.64 (0.37–1.09)	—	0.56	0.53 (0.28–1.03)	—	0.46	0.42 (0.25–0.96)	—	0.52	0.50 (0.26–0.95)
TCPy
Unadjusted	0.10	1.86	1.87 (1.29–2.67)	99.3	1.81	1.90 (1.04–3.22)	98.4	1.46	1.57 (0.89–2.62)	95.5	1.20	1.29 (0.73–2.26)
SG-adjusted	—	1.93	1.92 (1.31–2.66)	—	1.94	1.89 (1.21–3.14)	—	1.67	1.70 (1.06–2.65)	—	1.46	1.54 (0.95–2.49)
PNP
Unadjusted	0.25	2.30	2.35 (1.53–3.49)	99.7	2.05	2.12 (1.15–3.94)	99.5	1.73	1.79 (0.95–3.14)	100	1.79	1.71 (0.95–3.30)
SG-adjusted	—	2.36	2.34 (1.65–3.40)	—	2.20	2.20 (1.32–3.76)	—	1.98	2.02 (1.17–3.19)	—	2.19	2.18 (1.33–3.66)
mPYRs
3-PBA
Unadjusted	0.01	0.20	0.19 (0.11–0.38)	98.0	0.17	0.17 (0.08–0.36)	94.7	0.13	0.12 (0.05–0.30)	91.6	0.11	0.10 (0.05–0.27)
SG-adjusted	—	0.21	0.19 (0.12–0.35)	—	0.19	0.17 (0.09–0.34)	—	0.15	0.13 (0.07–0.27)	—	0.14	0.13 (0.06–0.28)
trans-DCCA
Unadjusted	0.02	0.25	0.24 (0.14–0.43)	95.6	0.21	0.21 (0.09–0.45)	92.9	0.14	0.15 (0.05–0.34)	93.0	0.14	0.12 (0.06–0.31)
SG-adjusted	—	0.25	0.23 (0.14–0.42)	—	0.24	0.23 (0.12–0.48)	—	0.15	0.14 (0.07–0.31)	—	0.14	0.13 (0.07–0.28)
mNNIs
IMI
Unadjusted	0.025	0.05	0.05 (0.03–0.09)	72.1	0.04	0.03 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.08$)	74.5	0.03	0.03 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.07$)	73.8	0.03	0.03 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.07$)
SG-adjusted	—	0.05	0.05 (0.03–0.10)	—	0.04	0.04 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.10$)	—	0.04	0.03 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.08$)	—	0.04	0.03 ($<\text{MDL\hspace{0.17em}to\hspace{0.17em}}0.07$)
IMI-olefin
Unadjusted	0.025	0.57	0.53 (0.31–1.01)	100	0.46	0.43 (0.21–0.93)	100	0.38	0.35 (0.18–0.79)	100	0.37	0.33 (0.16–0.83)
SG-adjusted	—	0.60	0.55 (0.34–1.01)	—	0.53	0.47 (0.25–1.05)	—	0.40	0.36 (0.19–0.77)	—	0.38	0.33 (0.16–0.79)
5-hydroxy-IMI
Unadjusted	0.05	0.97	0.93 (0.48–1.79)	99.9	0.79	0.73 (0.35–1.82)	99.8	0.57	0.54 (0.25–1.23)	100	0.55	0.52 (0.23–1.23)
SG-adjusted	—	1.00	0.92 (0.54–1.73)	—	0.92	0.81 (0.43–1.92)	—	0.60	0.56 (0.28–1.23)	—	0.56	0.50 (0.24–1.18)
DN-IMI
Unadjusted	0.025	0.13	0.11 (0.06–0.22)	84.2	0.11	0.09 (0.04–0.21)	82.0	0.08	0.06 (0.04–0.13)	84.0	0.09	0.07 (0.04–0.17)
SG-adjusted	—	0.14	0.12 (0.07–0.22)	—	0.12	0.11 (0.06–0.23)	—	0.08	0.07 (0.04–0.15)	—	0.09	0.08 (0.04–0.16)
DM-ACE
Unadjusted	0.01	1.17	1.13 (0.70–1.94)	99.9	0.96	0.93 (0.47–1.93)	99.9	0.80	0.79 (0.39–1.64)	100	0.81	0.75 (0.37–1.62)
SG-adjusted	—	1.25	1.20 (0.72–2.01)	—	1.11	1.07 (0.54–2.18)	—	0.84	0.83 (0.41–1.69)	—	0.83	0.77 (0.39–1.55)
THM
Unadjusted	0.025	0.12	0.11 (0.06–0.23)	95.4	0.08	0.06 (0.03–0.18)	96.3	0.08	0.07 (0.03–0.18)	95.8	0.08	0.07 (0.03–0.17)
SG-adjusted	—	0.13	0.12 (0.06–0.23)	—	0.09	0.08 (0.03–0.21)	—	0.08	0.07 (0.03–0.17)	—	0.08	0.07 (0.03–0.17)
CLO
Unadjusted	0.05	0.16	0.15 (0.09–0.27)	94.0	0.12	0.11 (0.05–0.23)	91.8	0.10	0.09 (0.05–0.20)	92.9	0.11	0.10 (0.05–0.23)
SG-adjusted	—	0.17	0.15 (0.10–0.28)	—	0.14	0.13 (0.07–0.27)	—	0.11	0.10 (0.05–0.20)	—	0.11	0.10 (0.05–0.21)
DM-CLO
Unadjusted	0.01	0.28	0.34 (0.17–0.58)	94.2	0.24	0.32 (0.14–0.59)	88.8	0.16	0.23 (0.07–0.42)	94.4	0.21	0.27 (0.12–0.48)
SG-adjusted	—	0.30	0.36 (0.18–0.63)	—	0.28	0.38 (0.17–0.67)	—	0.17	0.23 (0.09–0.46)	—	0.22	0.27 (0.11–0.53)

Note: Other target analytes (including DEDTP, DMDTP, 4F-3-PBA, ACE) were not listed here because they were not found in the samples or the detection frequency was $<30\%$. 25th and 75th means the 25th and 75th percentile concentrations. —, Not applicable; 3-PBA, 3-phenoxybenzoic acid; 4F-3-PBA, 4-fluoro-3-phenoxybenzoic acid; 5-hydroxy-IMI, 5-Hydroxy-imidacloprid; ACE, acetamiprid; CLO, clothianidin; DEDTP, diethyl dithiophosphate; DEP, diethyl phosphate; DETP, diethyl thiophosphate; DF, detection frequency; DM-ACE, desmethyl-acetamiprid; DM-CLO, desmethyl-clothianidin; DMDTP, dimethyl dithiophosphate; DMP, dimethyl phosphate; DMTP, dimethyl thiophosphate; DN-IMI, desnitro-imidacloprid; GM, geometric mean; IMI, imidacloprid; IMI-olefin, imidacloprid-olefin; MDL, method detection limit; mNNIs, metabolites of neonicotinoid insecticides; mOPPs, metabolites of organophosphate insecticides; mPYRs, metabolites of pyrethroids; PNP, para-nitrophenol; SG, urinary specific gravity; TCPy, 3,5,6-trichloro-2-pyridinol; THM, thiamethoxam; trans-DCCA, trans-3-(2,2-dichlorovinyl)-2,2-dimethylcyclopropanecarboxylic acid.

Moderate ($r=0.55–0.79$) to strong correlations ($r=0.86$ for IMI-olefin and 5-hydroxy-IMI) were observed among the metabolites of the same class insecticides based on the averaged concentrations of three trimesters (Figure S1 and Excel Table S2), and similar Spearman’s correlation coefficients were observed for trimester-specific concentrations (Figures S2–S4 and Excel Tables S3–S5). The correlations between metabolites of different classes are weak ($r<0.5$).

Single Chemical Analyses of the Associations between Individual Urinary Biomarkers and Children’s MDI/PDI Scores

MLR models using the averaged concentrations during pregnancy showed that significant associations with low Bayley scores were observed only among boys (Figure 1; Table S7). Specifically, each 1-unit increase in ln-transformed averaged urinary concentrations of DEP, DETP, TCPy, PNP, and trans-DCCA during pregnancy was significantly associated with a decrease of 3.79 points (95% CI: $-6.77$, $-0.81$), 2.94 points (95% CI: $-5.58$, $-0.29$), 5.51 points (95% CI: $-9.13$, $-1.90$), 4.50 points (95% CI: $-7.97$, $-1.02$), and 3.50 points (95% CI: $-5.78$, $-1.22$) respectively in boys’ MDI scores. After multiple testing correction, the associations of DEP ($p_{FDR}=0.04$), TCPy ($p_{FDR}=0.02$), PNP ($p_{FDR}=0.04$), and trans-DCCA ($p_{FDR}=0.02$) with lower boys’ MDI scores remained significant (Table S7). The interactions between child sex and DEP ($p_{sex\text{-}int}=0.04$), DETP ($p_{sex\text{-}int}=0.04$), TCPy ($p_{sex\text{-}int}=0.01$), PNP ($p_{sex\text{-}int}=0.05$), and trans-DCCA ($p_{sex\text{-}int}=0.01$) on Bayley scores were found to be significant (Table S7).

Figure 1.

Adjusted associations between the averaged concentrations (ng/mL, ln-transformed SG-adjusted) of maternal urinary mOPPs, mPYRs, and mNNIs over the three trimesters and children’s MDI and PDI scores using multivariate linear regression (MLR) models for participants from a birth cohort study in Wuhan, China, 2014–2017 ($N=\mathrm{1,041}$ participants, $n=\mathrm{3,123}$ samples). The models were adjusted for maternal age (categorical), PBMI (categorical), maternal education (categorical), passive smoking (categorical), folic acid supplementation during pregnancy (categorical), parity (categorical), child sex (categorical), breastfeeding duration (categorical), and delivery mode (categorical). Sex-stratified analyses were adjusted for all the abovementioned confounders except child sex. *$p_{FDR}<0.05$ were considered statistically significant (FDR corrections were performed to adjust for multiple comparisons). The numerical values are listed in Table S7. Note: 3-PBA, 3-phenoxybenzoic acid; 5-hydroxy-IMI, 5-hydroxy-imidacloprid; CLO, clothianidin; DEP, diethyl phthalate; DETP, diethyl thiophosphate; DM-ACE, desmethyl-acetamiprid; DM-CLO, desmethyl-clothianidin; DMP, dimethyl phosphate; DMTP, dimethylthiophosphate; DN-IMI, desnitro-imidacloprid; IMI, imidacloprid; IMI-olefin, imidacloprid-olefin; MDI, mental development index; mNNIs, metabolites of neonicotinoid insecticides; mOPPs, metabolites of organophosphate insecticides; mPYRs, metabolites of pyrethroids; PBMI, prepregnancy body mass index; PDI, psychomotor development index; PNP, 4-nitrophenol; TCPy, 3,5,6-trichloro-2-pyridinol; THM, thiamethoxam; trans-DCCA, trans-3-(2,2-dichlorovinyl)-2,2-dimethylcyclopropanecarboxylic acid.

In trimester-specific analyses based on GEE models, only the insecticide metabolites at the first trimester had significantly inverse associations with children’s MDI/PDI scores (Figure 2; Table S8). Among all the children, higher urinary concentrations of TCPy and trans-DCCA of the first trimester were significantly associated with lower MDI or both MDI and PDI. After stratification by child sex, significantly inverse associations were found only in boys (at the first trimester). Specifically, each 1-unit increase in ln-transformed maternal urinary concentrations of two mOPPs (TCPy and PNP), one mPYR (trans-DCCA), and one mNNI (THM) was significantly associated with a decrease of 3.16 points (95% CI: $-5.59$, $-0.74$), 3.06 points (95% CI: $-5.45$, $-0.68$), 2.24 points (95% CI: $-3.89$, $-0.58$), and 1.53 points (95% CI: $-2.92$, $-0.13$) respectively in boys’ MDI scores; each 1-unit increase in ln-transformed maternal urinary concentrations of two mPYRs (3-PBA and trans-DCCA) was significantly associated with a decrease of 1.55 points (95% CI: $-2.90$, $-0.20$) and 1.90 points (95% CI: $-3.16$, $-0.64$) respectively in boys’ PDI scores. After multiple testing correction, the associations of TCPy ($p_{FDR}=0.04$), PNP ($p_{FDR}=0.04$), and trans-DCCA ($p_{FDR}=0.04$) at the first trimester with lower MDI scores of boys remained significant (Table S8). At the first trimester, the interactions between child sex and PNP ($p_{sex\text{-}int}=0.05$) and trans-DCCA ($p_{sex\text{-}int}=0.04$) on Bayley scores were found to be significant (Table S8). No significantly inverse associations were observed at the second or third trimester after multiple testing correction (Table S8).

Figure 2.

Adjusted associations between trimester-specific concentrations (ng/mL, ln-transformed SG-adjusted) of maternal urinary mOPPs, mPYRs, and mNNIs over the three trimesters and children’s (A) MDI and (B) PDI scores using generalized estimating equations (GEE) models for participants from a birth cohort study in Wuhan, China, 2014–2017 ($N=\mathrm{1,041}$ participants, $n=\mathrm{1,041}$ samples for each specific trimester). The models were adjusted for maternal age (categorical), PBMI (categorical), maternal education (categorical), passive smoking (categorical), folic acid supplementation during pregnancy (categorical), parity (categorical), child sex (categorical), breastfeeding duration (categorical), delivery mode (categorical), and sampling seasons (categorical). Sex-stratified analyses were adjusted for all the abovementioned confounders except child sex. *$p_{FDR}<0.05$ were considered statistically significant (FDR corrections were performed to adjust for multiple comparisons). Details of chemical abbreviations are provided in Figure 1, and the numerical values are listed in Table S8. Note: FDR, false discovery rate; MDI, mental development index; mNNIs, metabolites of neonicotinoid insecticides; mOPPs, metabolites of organophosphate insecticides; mPYRs, metabolites of pyrethroids; PBMI, prepregnancy body mass index; PDI, psychomotor development index; $p_{FDR}<0.05$, $p$-value for false discovery rate; SG, specific gravity.

In the sensitivity analyses that a) additionally adjusted for more covariates, b) excluded women with gestational diabetes or gestational hypertension, and c) excluded children with preterm birth or low birth weight in MLR models, most significant associations between insecticide biomarkers (especially for DEP, TCPy, PNP, trans-DCCA, and DM-CLO) and Bayley scores remained (Excel Tables S6–S8). Although the significant positive associations between CLO and Bayley scores among all the children disappeared in the three sensitivity analyses (Excel Tables S6–S8), some significantly positive associations of maternal urinary biomarker concentrations with Bayley scores were observed (Figures 1–2; Tables S7–S8). Specifically, after multiple testing correction, higher averaged and trimester-specific concentrations of DM-CLO were significantly associated with higher MDI scores among the boys (Tables S7–S8), and higher concentration of DMP in the third trimester was also significantly associated with higher PDI scores among the girls (Table S8).

WQSR Analysis for the Mixture Effect

Based on the results of MLR and GEE models, boys might be more susceptible to insecticide exposure than girls, and the first trimester appeared to be a critical window of the susceptibility to insecticide exposures. For such findings with statistical significance, the WQSR analyses were further performed in the negative direction to evaluate the mixture effect of exposure to the selected insecticides on the Bayley scores. At the first trimester, a significant association was observed between a 1-quartile increase in the WQS index and lower MDI scores in boys ($\mathrm{\beta }=-3.02$ points; 95% CI: $-5.47$, $-0.57$), and trans-DCCA (18.2%) contributed the most to the mixture effect, followed by THM (16.5%), TCPy (12.6%), DN-IMI (12.1%), PNP (9.6%), DEP (7.0%), and 3-PBA (6.5%) (Figure 3; Table S9). For the averaged concentrations of the three trimesters, the WQS index was also associated with lower boys’ MDI scores ($\mathrm{\beta }=-2.80$ points; 95% CI: $-4.96$, $-0.64$), and TCPy made the largest contribution to the association (28.2%), followed by trans-DCCA (12.0%), PNP (11.7%), DEP (10.1%), and 3-PBA (9.4%). We also analyzed the association between the insecticide biomarker mixture and boy’s PDI scores, and found no statistical significance (Table S10).

Figure 3.

Associations between the maternal urinary biomarker mixture and boy’s MDI scores assessed with repeated holdout random subset WQSR in the negative direction, based on the concentrations (ng/mL) in the first trimester ($N=553$ participants, $n=553$ samples) and the averaged concentrations (ng/mL) over the three trimesters ($N=553$ participants, $n=\mathrm{1,659}$ samples) for participants from a birth cohort study in Wuhan, China, 2014–2017. The models were adjusted for maternal age (categorical), PBMI (categorical), maternal education (categorical), passive smoking (categorical), folic acid supplementation during pregnancy (categorical), parity (categorical), breastfeeding duration (categorical), delivery mode (categorical), and sampling seasons (except for average concentrations). *$p<0.05$ were considered statistically significant. Details of chemical abbreviations are provided in Figure 1, and the numerical values are listed in Table S9. Note: CI, confidence interval; MDI, mental development index; PBMI, prepregnancy body mass index; WQSR, weighted quantile sum regression.

Discussion

This study investigated the associations of maternal coexposure to OPPs, PYRs, and NNIs during pregnancy with child neurodevelopment. We found that a) the maternal urinary concentrations of some mOPPs, mPYRs, and mNNIs were significantly associated with lower neurodevelopment scores in children; b) the inverse associations of insecticide concentrations measured in the first trimester appeared to be more pronounced than the associations with second and third trimester measurements; c) the associations in boys were larger than those in girls; and d) trans-DCCA and TCPy were the major biomarkers associated with the mixture effect of the insecticide mixture exposure.

In addition, we compared biomarker concentrations in this study with those in other studies that investigated the associations of prenatal exposure to OPPs and PYRs with child neurodevelopment (Table S11). Maternal urinary biomarker concentrations of OPPs and PYRs observed in our study were generally within the same order of magnitude with those reported previously (Table S11). We also compared them with the data of women ($>19\text{\hspace{0.17em}years}$ of age) from NHANES in 2013–2014⁵³ or in 2017–2018⁵⁴ and found that pregnant women in this study had higher median concentrations of TCPy (1.29–1.90 vs. $0.84\text{\hspace{0.17em}ng}/\mathrm{mL}$), PNP (1.71–2.12 vs. $0.62\text{\hspace{0.17em}ng}/\mathrm{mL}$), DEP (2.70–3.75 vs. $1.84\text{\hspace{0.17em}ng}/\mathrm{mL}$), DETP (1.18–1.72 vs. $0.13\text{\hspace{0.17em}ng}/\mathrm{mL}$), and DMP (2.99–4.74 vs. $1.15\text{\hspace{0.17em}ng}/\mathrm{mL}$) but lower median concentrations of 3-PBA (0.10–0.17 vs. $0.62\text{\hspace{0.17em}ng}/\mathrm{mL}$) and DMTP ($0.38–0.46$ vs. $0.54\text{\hspace{0.17em}ng}/\mathrm{mL}$) compared with women from the United States (Table S11). So far, few studies have reported the associations between biomarkers of NNI exposure during pregnancy and child neurodevelopment. Nevertheless, a few studies provided data on mNNIs in pregnant women; for example, Anai et al.⁶² observed that the urinary median concentrations of THM ($7.4\mathrm{\mu }\mathrm{g}/\mathrm{g}$ creatinine; LOD: $1.03\text{\hspace{0.17em}ng}/\mathrm{mL}$) and CLO ($15.3\mathrm{\mu }\mathrm{g}/\mathrm{g}$ creatinine; LOD: $0.43\text{\hspace{0.17em}ng}/\mathrm{mL}$) in pregnant women from Japan were higher than those in this study (CLO: $0.09–0.11\text{\hspace{0.17em}ng}/\mathrm{mL}$; THM: $0.06–0.07\text{\hspace{0.17em}ng}/\mathrm{mL}$).

The results from this study suggest that, compared with the second or third trimester, the first trimester may be a more sensitive window of exposure to the insecticides in association with Bayley scores among boys. The first trimester is the most critical period of fetal brain development,⁶³ and this may explain what we observed in this study. van den Dries et al.⁶⁴ also found that early pregnancy was the sensitive window of phthalate exposure in association with neurodevelopment in the Generation R study. Nevertheless, very few studies have examined the sensitive window for the relationship between prenatal insecticide exposure and Bayley scores (Table S11). Qi et al.⁶⁵ found an association of 3-PBA (trans-DCCA was not measured) at the second trimester (vs. the first and third trimesters) with decreased Bayley-III scores of infants at 1 year of age ($n=419$). However, Jusko et al.²² found an association of elevated DAPs at the third trimester (vs. the first and second trimesters) with lower child intelligence quotient (IQ) at 7 years of age ($n=708$) in the Generation R study. This inconsistency may be attributed to differences in biomarker selection, sample size, exposure level, neurodevelopment outcome assessed, child age of neurodevelopment assessment, and exposure profiles of insecticides. Our findings of trimester-specific analysis may provide clues for future studies and health risk interventions related to pesticide exposure.

The Bayley PDI or MDI scores of the boys were slightly lower than the girls at the same age in this study (Table 1), which was generally consistent with a very recent study in Shanghai, China.⁶⁶ We further observed stronger associations between the higher concentrations of the insecticide metabolites and lower MDI scores for the boys. Such associations were not significant in the girls, possibly owing to the (well-documented) resilience of the female brain to neurotoxic injury caused by contaminants.⁶⁷ Estrogen itself is neuroprotective,⁶⁷ and estrogen-mediated receptors can activate multiple molecular neuroprotective and neuro-reparative responses.^67,68 Interference with development of sex-dimorphic neuroendocrine pathways (e.g., sex-specific expression level of estrogen receptors in the brain),^68,69 neuroepigenetics,⁷⁰ thyroid hormone homeostasis,^71,72 neuropeptide and neurotransmitter signaling,^73,74 and neuroinflammation³⁴ have been suggested to be responsible for such sex-specific neurotoxic effects of environmental chemicals like OPPs. Several previous epidemiological studies have also found similar sex-specific associations of prenatal exposure to OPPs (based on biomonitoring data of urinary DAPs^13,21,23,75 or chlorpyrifos in cord blood⁷⁶) with adverse neurodevelopment in boys. Such sex-specific findings for prenatal exposure to PYRs and NNIs have not been reported in human beings by others yet, and further confirmation is needed. Nevertheless, the sex-specific effects of PYRs and NNIs on neurotoxicity have been characterized in animal models. For instance, prenatal exposure of mice to deltamethrin ($3\mathrm{mg}/\mathrm{kg}$) induced behavioral abnormalities only in male offspring,⁷⁷ and higher striatal dopamine transporter protein levels and D1 dopamine receptor levels in male offspring may be responsible for the differed neurodevelopment changes.⁷⁷ In another animal study, conducted by Kubo et al.,⁷⁸ administration of CLO at the chronic no observed adverse effect level (NOAEL) dose ($50\mathrm{mg}/\mathrm{kg}$) reduced locomotor activities and elevated anxiety-like behaviors more in male mice than in females. They suggested that male-dominant changes in neuroactivities in several brain regions (including the paraventricular thalamic nucleus and the dentate gyrus) could explain the sex-related differences in neurotoxicity induced by CLO.⁷⁸

Single chemical analyses in this study showed the significant associations of higher maternal urinary concentrations of certain mOPPs (i.e., DEP, TCPy, and PNP) and mPYRs (i.e., trans-DCCA) with lower MDI scores. For each 1-unit increase in ln-transformed concentrations of these metabolites, the Bayley scores decreased by 2–4 points. In previous studies that explored the associations between pesticide exposure during pregnancy and Bayley scores in children (Table S11), some studies also found that prenatal exposure to OPPs (e.g., assessed using the averaged urinary concentrations of TCPy and DAPs at the first and second trimesters,¹¹ urinary DAPs at the third trimester,^20,79 or chlorpyrifos in cord blood^19,80) or PYRs (assessed using urinary mPYRs prior to delivery,⁸¹ at the first and second trimesters,⁶⁵ or at the third trimester⁸²) were associated with poorer Bayley scores of children (2–36 months of age). In these studies, for each 1-unit,^65,82 an interquartile range,⁷⁹ or 10-fold increase^11,20,81 in urinary insecticide biomarkers during pregnancy, the children’s Bayley scores decreased by 0.03–4.42 points. A couple of studies categorized the exposure levels into high and low, and found higher chlorpyrifos exposure ($>6.17\text{\hspace{0.17em}pg}/\mathrm{g}$) produced an $\sim 3.0\text{- to}7.0$-point decrease in Bayley scores compared with those with lower exposure ($\le 6.17\text{\hspace{0.17em}pg}/\mathrm{g}$)^19,80 (Table S11). However, Donauer et al.⁸³ (using maternal urinary DAPs at gestational weeks 16 and 26) and Watkins et al.²⁵ (using maternal urinary 3-PBA at the third trimester) did not find such associations.

In addition to Bayley scores, some studies (Table S11) have shown that prenatal exposure to OPPs or PYRs or both of them were associated with other neurodevelopment outcomes of children [e.g., developmental quotients,^13,21,76 IQ scores,^22,23,84,85 neurobehavioral development of neonates,⁸⁶ traits related to ASD,^87,88 traits related to attention deficit hyperactivity disorder,^31,89 decreased brain activity⁹⁰]. However, other studies did not find such associations of childhood outcomes with prenatal exposure to OPPs,^91–93 PYRs,^24,94,95 or both.²⁶ Differences in sample size, time point of sample collection, biomarkers selected, child age at examination, exposure levels, exposure characteristics, and individual susceptibility might partly account for the inconsistency across these studies.

We also examined the association of maternal urinary concentrations of PNP and mNNIs during pregnancy with child neurodevelopment. Urinary PNP has been recommended as a biomarker of exposure to parathion and methyl parathion.⁹⁶ Consistent with that reported by Li and Kannan,⁴⁰ PNP was detected with high DF (99.5%–100%) in this study. However, parathion and methyl parathion have been banned in China since 2007 owing to their high toxicity.⁹⁷ Meanwhile, PNP can be a metabolite of itself^98,99 and nitrobenzene.^100,101 Therefore, the concentrations of PNP do not necessarily represent OPP exposure, and exposure sources of PNP parent compounds for the general population merit further studies. In addition, PNP,⁹⁹ nitrobenzene,¹⁰⁰ parathion,¹⁰² and methyl parathion¹⁰³ have all been demonstrated to have neurotoxicity, and their neurodevelopmental toxicities need further studies.

In the present study, we found some significantly positive associations of DM-CLO or DMP with Bayley scores, which might be confounded by some unmeasured protective factors such as those from the consumption of fruits and vegetables.^22,104 Inconsistent with our study, three previous studies from California have identified potential relationships between prenatal residential proximity to agricultural use of neurotoxic pesticides (including IMI) and an increased risk of neural tube defects¹⁰⁵ or lower IQ scores in 7-y-old children.^17,18 In China, the general population is exposed to both traditional insecticides (e.g., OPPs and PYRs) and new emerging insecticides (e.g., NNIs) mainly through the diet, but residue concentrations differ among various foodstuffs. Specifically, CLO (the parent compound of DM-CLO) is mainly ingested through fruits and vegetables in China,²⁹ in contrast to the traditional insecticide chlorpyrifos¹⁰⁶ (the parent compound of TCPy, DETP, and DEP).^107,108 Such comparison data were not available on dichlorvos (a major parent insecticide of DMP),¹⁰⁸ malathion (a major parent insecticide of DMP and DMTP),¹⁰⁸ and cypermethrin/permethrin (the major parent insecticides of trans-DCCA),⁸² so we had to resort to data on their registration amounts in China. According to registration information of pesticides¹⁰⁹ in China, dichlorvos (62.8%) and malathion (57.1%) are mainly registered for use on fruits and vegetables,¹⁰⁹ whereas cypermethrin/permethrin are mainly registered as a sanitary insecticide (e.g., for mosquito control) and for use on cotton (55.3%) rather than on fruits and vegetables (31.6%).¹⁰⁹ Thus, the nutritional benefits of fruits and vegetables may counteract or outweigh the adverse effects of CLO/DM-CLO and dichlorvos/malathion/DMP on child neurodevelopment. On the other hand, we cannot rule out the influence of some other unmeasured confounding factors of socioeconomic status and dietary habits that may mask the adverse effects of such insecticides.²⁶ The reason why higher DMP levels in the third trimester were associated with better PDI scores only in girls is unclear; more studies are warranted to understand whether the underlying mechanisms are related to estrogen and its receptors.

Based on the WQSR model, which can help reduce the likelihood of collinearity, model instability, and the reversal paradox,⁵⁷ we found that higher insecticide mixture exposure levels were significantly related to lower MDI scores among the boys, with trans-DCCA and TCPy identified as the major biomarkers of concern. Similarly, in a mixture analysis (i.e., WQSR) of prenatal exposure to 26 endocrine-disrupting chemicals (EDCs, including TCPy and 3-PBA but not trans-DCCA) measured in the first trimester urine or blood, Tanner et al.⁵⁸ observed an inverse association of EDC mixture with IQ scores of 7-y-old boys, and they identified TCPy and 3-PBA as the major biomarkers of concern. Another study observed that prenatal exposure (using averaged concentrations at $<18$, $18–25$, and $>25\mathrm{wk}$ of gestation) to selected nonpersistent chemical mixtures [assessed by measurements of phthalate metabolites, bisphenol A, and five nonspecific mOPPs (DEP, DETP, DMP, DMTP, and DMDTP) in urine] was associated with lower nonverbal IQ in 6-y-old children,¹¹⁰ although they found the association was mainly driven by a phthalate mixture.

Although trans-DCCA and TCPy were detected with lower concentrations compared with other biomarkers (e.g., DMP and DEP), they were identified as the major biomarkers of concern affecting neurodevelopment. First, this finding may be attributed to the much higher neurotoxicity of their major parent insecticides (cypermethrin and chlorpyrifos, respectively) than the major parent insecticides of DMP¹⁰⁸ (dichlorvos and malathion)¹¹¹ applied in China,^109,112 although available toxicological data did not contain sufficient information on the effects on the cognitive domain (learning and memory).¹¹³ For example, concerning functional alteration of the motor division of rodent nervous system, the NOAEL values of cypermethrin and chlorpyrifos [both are $5.0\mathrm{mg}/\mathrm{kg}$ per body weight (BW) per day]¹¹¹ were $\sim 30$ times lower than that of malathion [($\text{NOAEL}=\mathrm{1,486}\mathrm{mg}/\mathrm{kg}$ per BW per day),¹¹¹ which is a major parent compound of DMP and DMTP¹⁰⁸ applied in China¹⁰⁹]. Second, DAPs are nonspecific metabolites of OPPs,¹⁰⁸ and the related NOAEL values differed largely among various OPPs [for example, the NOAEL value is 17 and $0.1\mathrm{mg}/\mathrm{kg}$ per BW per day for malathion and dimethoate, respectively, concerning the effect of acetylcholinesterase inhibition,¹¹¹ and malathion is frequently used in China but dimethoate is not,¹⁰⁹ despite both being parent insecticides of DMP and DMTP¹⁰⁸], thus, it would be hard to examine/interpret their effects only on the basis of the concentrations of the nonspecific DAPs (without knowing the exposure levels of the specific parent OPPs). Third, some OPPs would be transformed rapidly into DAPs after agricultural application [e.g., dichlorvos has a very short half-life (4.6–6.8 h) in beans, potato, and tomato, being transformed into DMP, methyl phosphate, and others¹¹⁴]; thus DAP residues (which are not considered toxic¹⁰⁸) in fruits and vegetables may confound exposure biomonitoring and risk assessment.¹¹⁵ Last, DMP and DEP could also be metabolized from TMP and TEP,⁵¹ which are not insecticides but are instead organophosphate esters (another class of widespread environment contaminants),⁵² and TMP¹¹⁶ and TEP¹¹⁷ are neurotoxic only at high doses. This may also make it hard to interpret the effects related to DMP and DEP.

The well-known toxicity of OPPs occurs via inhibition of cholinesterase, but in recent years the detrimental effects of exposure to low-dose OPPs have been attributed to some noncholinergic mechanisms,¹¹⁸ such as exacerbated oxidative stress¹¹⁹ and disruption of thyroid hormone homeostasis.¹²⁰ The neurotoxic mechanisms of cypermethrin were hypothesized to include perturbation of the dopamine system⁷⁴ and increase of neuroinflammation.¹²¹ It is also noteworthy that chlorpyrifos has been gradually banned in many countries owing to its reported risks to nontarget organisms, particularly developmental neurotoxicity in children,^122,123 but is still widely used in China. Similarly, some isomers of cypermethrin (including alpha-cypermethrin, beta-cypermethrin, and theta-cypermethrin) are not approved for use in the European Union anymore,¹²⁴ owing to their developmental neurotoxicity in rats [range of acceptable daily intake values for those isomers: $0.00125–0.0016\mathrm{mg}/\mathrm{kg}$ BW per day].¹²⁵ Permethrin is not approved for use in the European Union either.¹²⁴ It is the right time for China to further regulate them and other parent insecticides of TCPy and tran-DCCA, as well as THM.

The strength of this study was analyzing the mixture effect of multiple insecticides on child neurodevelopment with a relatively large sample size and repeated measurements across trimesters. However, some limitations should be noted. First of all, we did not collect information on dietary patterns of the pregnant women, which may cause some potential bias of our findings. Further, the confounding effects of other pollutants were not considered in this work. Finally, we did not assess the associations between Bayley scores and children’s postnatal exposure to these insecticides, which might also have some impact. Future prospective studies are warranted to confirm our findings by considering the abovementioned limitations.

Conclusions

Our findings suggest that maternal insecticide exposure in the first trimester may compromise child neurodevelopment, and the associations are particularly pronounced in boys. In addition, we found that trans-DCCA and TCPy contributed the most to the mixture effect. In addition to those factors that are commonly recognized to affect child neurodevelopment (e.g., nutrition in pregnancy, socioeconomic status, genetics), our findings highlight the necessity for public health and policy measures to reduce pesticide exposures among pregnant women and warrant additional investigations.